Development and Validation of a Cuproptosis-Related Prognostic Model for Acute Myeloid Leukemia Patients
A new study published in Scientific Reports has developed a cuproptosis-related prognostic model for acute myeloid leukemia (AML) patients using machine learning with stacking. The research team downloaded RNA-seq data of 151 patients and corresponding clinical information from the TCGA-LAML database, as well as data from two additional datasets from the GEO database. After excluding samples without complete survival information, a total of 136 AML patients were included from the TCGA-LAML database, while 553 and 240 patients were included from the GEO database.
The researchers adjusted the batch effects in the data using the normalizeBetweenArrays function of the limma R package. They then reviewed 10 cuproptosis-related genes (CRGs) from the literature and used Spearman rank correlation analysis to identify RNA related to cuproptosis. A univariate Cox hazard regression analysis was performed using the GSE37642 dataset to identify CRGs associated with AML survival.
The performance of the model was evaluated in terms of discrimination and calibration. Discrimination refers to the ability of the model to distinguish between patients at different risks. The researchers plotted cumulative/dynamic time-dependent receiver operating characteristic (ROC) curves at 1, 2, and 3 years using the survivalROC package. The area under the ROC curve (AUC) was used to express the discriminatory power of the model over different time scales. They also conducted Kaplan-Meier survival analysis to assess the prognostic value of the linear predictor.
Calibration, on the other hand, measures the agreement between the predicted risk of death and the observed risk of death. Calibration plots were generated using the rms package to assess the calibration of the Cox hazards model. The researchers also examined the correlation between cuproptosis-related linear predictors and therapeutic targets using Pearson correlation analysis. Additionally, they predicted chemotherapy drugs based on the Cancer Drug Sensitivity Genomics (GDSC) database and identified potential drug targets using the Drug-Gene Interaction database (DGIdb).
To develop a survival prediction model, four machine learning algorithms were utilized: random survival forest (RSF), survival support vector machine (survival-SVM), generalized boosted regression modeling (GBM), and eXtreme Gradient Boosting (XGBoost). The algorithms were applied using the stacking linear predictor and clinical factors. The researchers then compared the differences in risk among different AML groups.
The findings of this study have the potential to improve the prognostic assessment and treatment strategies for AML patients. The cuproptosis-related prognostic model developed through machine learning approaches could provide valuable insights into patient outcomes and help guide personalized treatment decisions. Further research and validation are needed to fully understand the clinical implications of this model.
